A Methodology for Testing the Accuracy of Yield Predictions from Weather‐Yield Regression Models for Corn1

Abstract

AbstractReliable crop‐weather‐technology models are needed for yield prediction and for the determination of climatic risks in crop production. Accurate and timely assessment of crop yields would help both government and industry allocate the food supply more efficiently. Analysis of variance (ANOVA) techniques were used to evaluate the accuracy of yield predictions for corn (Zea mays L.) in Indiana counties with four statistical models: 1) Thompson approach 2) modified Thompson approach, with a nitrogen technology term, 3) a 1974 model by Leeper, Runge, and Walker and 4) a 1975 model by Dale and Hodges. For models 1), 2), and 4), separate versions were developed with the particular series of weather and corn yield data for each county. All models were used to predict yearly average corn yields for a county with data not used to fit the regression coefficients. A significant difference among models was detected by the ANOVA. Multiple comparison tests indicated that regression models 2), 3), and 4) were more accurate than model 1). This result was attributed mainly to the handling of technology or the weather‐technology interaction effects on corn yields. The trend variables used in model 1) acted to confound weather and technology effects and create unstable regression models. Although the direct application of the results of this study is limited to Indiana, the general regression problems and methodology for testing regression models for their accuracyin predicting crop yields are believed to have universal application.